Parallelization of robust multigrid technique

Abstract

A new approach to parallelization of multigrid algorithms for solving elliptic partial differential equations is proposed. The approach is based on the mutual adaption of multiprocessor architecture and the cycle of the robust multigrid technique. Adaption of multiprocessor architecture to the multigrid technique gives an optimum load balance between the processors. On the other hand, adaption of multigrid cycle to the multiprocessor architecture minimizes data transport between processors. Parallelization of the robust multigrid technique on coarse levels is smoother-independent. The paper shows that parallel implementation of intricate problems gives higher degree of parallelism as compared with the simple ones.